摘要
针对行人复杂多变的运动形式给室内定位带来较大偏差的问题,提出了一种基于加速度时域特征的行人运动分类方法,并利用分类结果进行室内行人三维定位。利用垂直加速度的变化规律将加速度信号划分为连续的单步信号,计算单步周期内加速度信号的时域特征,基于BP神经网络和支持向量机设计一种二分树结构的分类器。经大量人员运动实验验证,该分类器对走、跑,上下楼梯3类运动形式的分类准确率接近100%,上、下楼梯的分类准确率为95%;在行人运动形式确定的情况下,利用不同的步长模型和航向信息进行室内三维定位,定位误差为1.5 m。
The moving form of the pedestrian is complicated and changeable,which brings great deviation to the real time positioning. A pedestrian motion classification method based on acceleration time domain feature is proposed. The vertical acceleration signal is divided into continuous single-step signal. The time domain features are used to train the BP neural network classifier and SVM classifier.In the end,a motion classifier based on dichotomy tree is proposed. Through a large number of movement experimental verification,the classification accuracy rate among walk,run and up / down stair is nearly 100%,and the classification accuracy rate between up-stair and down-stair is about 95%.Different step-length models and headings are used to calculate the position information,with the tracking error at 1. 5 m.
出处
《北京信息科技大学学报（自然科学版）》
2016年第5期82-86,96,共6页
Journal of Beijing Information Science and Technology University
基金
国家自然科学基金资助项目(61471046)
北京市教委市属高校创新能力提升计划项目(TJSHG201510772017)
高动态导航技术北京市重点实验室开放课题
关键词
室内定位
运动分类
时域特征
神经网络
支持向量机
indoor positioning
motion classification
time-domain feature
neural network
support vector machine
作者简介
赵辉，男，硕士研究生；
通讯作者：李擎，女，博士，教授。
